Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 1062967.6168812835,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 75592.69360516561,
-      "required_pages_per_second_upper_estimate": 0.006552183093765394
+      "required_token_positions_per_second": 75827.27794091449,
+      "required_pages_per_second_upper_estimate": 0.006572516269439544
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 1206672.263377472,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9449.086700645701,
-      "required_pages_per_second_upper_estimate": 0.0037693819613234805
+      "required_token_positions_per_second": 9478.40974261431,
+      "required_pages_per_second_upper_estimate": 0.0037810793611833056
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 410042.7581134458,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 18898.173401291402,
-      "required_pages_per_second_upper_estimate": 0.004371759819896484
+      "required_token_positions_per_second": 18956.81948522862,
+      "required_pages_per_second_upper_estimate": 0.004385326559279534
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 2163486.5624068812,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28347.2601019371,
-      "required_pages_per_second_upper_estimate": 0.0017261994136284103
+      "required_token_positions_per_second": 28435.229227842934,
+      "required_pages_per_second_upper_estimate": 0.0017315562718577414
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 1546694.2737633225,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28347.2601019371,
-      "required_pages_per_second_upper_estimate": 0.0021822505528853986
+      "required_token_positions_per_second": 28435.229227842934,
+      "required_pages_per_second_upper_estimate": 0.0021890226597117562
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28347.2601019371,
-      "required_pages_per_second_upper_estimate": 0.3519780982894458
+      "required_token_positions_per_second": 28435.229227842934,
+      "required_pages_per_second_upper_estimate": 0.353070380419471
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.24157342121101522,
-    "producer_p95_seconds": 0.2214492290004273,
-    "gpu_update_seconds": 0.9753323566573082,
-    "producer_batches_per_second": 4.139528243574845,
-    "gpu_batches_per_second": 1.0252915256777018,
-    "producer_headroom": 4.037415837255342,
+    "producer_batch_seconds": 0.11149949907029111,
+    "producer_p95_seconds": 0.2128487860009045,
+    "gpu_update_seconds": 0.9723150032821926,
+    "producer_batches_per_second": 8.968650158415363,
+    "gpu_batches_per_second": 1.028473279363532,
+    "producer_headroom": 8.720353108216472,
     "sustainable": true,
-    "buffer_seconds": 60.47060611275311,
+    "buffer_seconds": 60.283530203495935,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 394246.8104229092,
-    "upload_bytes_per_second": 5948043.172309496,
+    "generation_bytes_per_second": 392540.82973729656,
+    "upload_bytes_per_second": 5941758.390667052,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 84.59618792071778,
+    "estimated_full_shards_per_hour": 84.23012417878316,
     "selection_policy": "Smallest shard within 90% of measured local peak, enlarged to amortize 30 seconds of measured page production; bounded to 256 MiB."
   },
   "upload_measurements": [
@@ -556,6 +556,12 @@
       "bytes": 10630071,
       "pages": 35,
       "shards": 3
+    },
+    {
+      "seconds": 1.0585477320000791,
+      "bytes": 5855284,
+      "pages": 31,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790519810984249790-cb9c9513/train.jsonl

@@ -123,3 +123,4 @@
 {"step": 127300, "loss": 0.6228926777839661, "lr": 0.0003, "grad_norm": 0.14142022313523672, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 18629941283, "tokens_per_second": 150273.28461330765, "data_wait_s": 1.1657000868581235e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 127400, "loss": 0.6144789196550846, "lr": 0.0003, "grad_norm": 0.14685550382924611, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 18644665520, "tokens_per_second": 150107.67555962576, "data_wait_s": 1.3332999515114352e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 127500, "loss": 0.6103075612336397, "lr": 0.0003, "grad_norm": 0.14554575553449317, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 18659408023, "tokens_per_second": 150570.71538785714, "data_wait_s": 1.6653000784572214e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 127600, "loss": 0.6087887566536665, "lr": 0.0003, "grad_norm": 0.1620604392941756, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 18674041976, "tokens_per_second": 149069.86359597286, "data_wait_s": 9.386001693201251e-06, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790519810.cad0b19a-59f2-43f0-8fb4-2a94097843e2-tjh7c.1336.0.1790519810984249790-cb9c9513

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:b17d132f48260add08f6ffc72c21f40d675c75bec19c9aa962b554b37dbd25aa
-size 266481
+oid sha256:4c5b20e8076ad437ef5b8a9b29b5382a7f222c31c22759d1b042e8fef64c6763
+size 268599